Comparisons as Predictable as the Sunrise

Russell Samora and Shelly Tan turn a corpus of English fiction into a lively anatomy of the classic 'as something as something' simile. Tiny distributions expose dependable clichés, comparison nouns split into specialists and generalists, and a candid methods section explains why the dataset is useful in aggregate rather than flawless sentence by sentence.

What is Comparisons as Predictable as the Sunrise?

The piece begins by making you finish a familiar comparison, then shows how your instinct sits beside thousands of writers' choices. Its central charts treat every adjective like a fingerprint: one towering bar signals a dominant cliché, while a gentler slope suggests more imaginative range. The perspective then flips from adjectives to nouns. Cucumber mostly stays cool, cats carry many qualities, stone is defined by what it lacks, and hell behaves less like a place than an all-purpose amplifier. Shelly Tan's bookish illustrations keep the statistical tour loose on its feet, while Russell Samora's methodology owns the hard edges of the extraction problem, including filtered one-offs and imperfect sentence-level labels.

What you can do there

• Complete an opening simile prompt
• Compare cliché distributions across adjectives
• Browse specialist, generalist, and ironic comparison nouns
• Read the extraction and normalization methods

Why we picked it

This is corpus analysis with ink on its cuffs. A rigid grammatical pattern becomes a surprisingly expressive map of writers' habits, and the specialist-versus-generalist framing makes a diversity index understandable without sanding away the oddball examples that make language fun.

How to get the most from it

Answer the opening blank before seeing the distribution, then compare the spikiest adjective charts with the flatter ones. In the noun sections, notice whether a comparison earns its meaning from a physical trait or merely intensifies the sentence. Finish with Data and Methods before treating any rank as exact.

Good to know

The study covers one parse-friendly simile form, not figurative language as a whole. It focuses on 500 frequent adjectives, filters many structural and literal matches, and conservatively groups related noun phrases. The author says the resulting dataset is dependable in aggregate but imperfect at sentence level; the corpus and calculations were not independently reproduced here.

Who made it, and when?

Russell Samora is the credited creator or organization. The earliest supported launch date we found is May 2026.